Microsoft is testing a Windows where users describe the software they need and Copilot generates it, but the vision currently depends on expensive hardware, hybrid cloud routing and new safeguards for AI-built applications.
Microsoft is pushing a different vision for Windows: not a marketplace of finished apps, but a system where software can be requested and generated on demand. After the company’s Windows and Surface event on 7 October, chief executive Satya Nadella said on X that every PC could become a kind of “infinite software factory”, with Copilot acting as the entry point rather than the app store. The idea is to move users away from browsing downloads and towards describing what they need in plain language. According to Microsoft, that request can now produce a native Windows application in a single step.
At the centre of the proposal is MAI-Code-1.1-Flash, Microsoft’s in-house coding model, which the company says has been designed for local hardware as well as cloud use. Microsoft describes it as a 137-billion-parameter mixture-of-experts system with 6.8 billion active parameters per token and a 256,000-token context window. To make it practical on personal devices, the model has been compressed with 3-bit quantisation, a technique that cuts memory use sharply while preserving the long context window. Microsoft says the model can be downloaded and run locally, with experimental access in GitHub Copilot expected by the end of the month.
The hardware demands remain substantial. Microsoft recommends machines with more than 120GB of RAM for the best experience, and third-party reporting based on Microsoft’s figures puts the model footprint at about 53GB, rising to roughly 75.5GB of memory use at full context. That means the feature is, for now, aimed at a very small class of PCs, not mainstream laptops. Microsoft’s own material says the local version is intended to reduce dependency on cloud inference, but the economics only work if customers are willing to absorb the cost of high-end memory and accelerators rather than pay per request in the cloud.
Even so, local execution does not mean a completely offline experience. Microsoft’s documentation indicates that task routing is handled dynamically, with GitHub HydraFusion deciding whether a request should run on the device or be sent to the cloud. That means Copilot’s local model is part of a hybrid system, not a sealed workstation tool. Microsoft also says the feature will reach users first as an experimental preview, reinforcing that the claims being made today are still subject to the limits of testing, policy controls and future updates.
The wider challenge is not only technical. Enterprise buyers remain cautious about software assembled by AI, especially when it touches business data or internal workflows. Retool said in a June 2026 report that 93% of IT leaders worry about AI-generated tools reaching production, while a Harris Poll found that 81% of chief information officers fear data leaks from software built by non-developers. Gartner has also projected that more than 40% of agentic AI projects will be abandoned by the end of 2027. Microsoft is trying to answer those concerns with MXC execution containers and policy controls in GitHub, but the real issue is maintenance: companies do not just buy features, they buy accountability, support and long-term compatibility.
The strategic logic is clear. If Copilot becomes the primary interface, Microsoft does not need to win the app-store battle in the traditional sense; it only needs to control the point of entry, the distribution path and, eventually, the commercial relationship with the user. That makes the move as much about platform power as software generation. Apple has already taken a different line. According to reporting cited by Macitynet and The Information, it has tightened rules around so-called vibe coding tools on iPhone and iPad, while still allowing AI assistance inside Xcode and a more supervised development pipeline. Microsoft is betting on software creation through an assistant. Apple is betting on orchestration of existing apps. The long-running struggle over who controls the user is simply shifting to a new layer.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
